CN111752733B

CN111752733B - Anomaly detection in a pneumatic system

Info

Publication number: CN111752733B
Application number: CN202010223367.5A
Authority: CN
Inventors: D·克劳斯; T·施特雷彻特; W·高赫尔
Original assignee: Festo SE and Co KG
Current assignee: Festo SE and Co KG
Priority date: 2019-03-29
Filing date: 2020-03-26
Publication date: 2024-03-08
Anticipated expiration: 2040-03-26
Also published as: DE102019108268A1; US20200310405A1; US11300952B2; CN111752733A

Abstract

The invention relates to error detection and localization in a pneumatic system (AA), in particular to an error detection module (FM), comprising: a read-in interface (I1) for reading in digital signals from the automation device (AA); -a first processor unit (P1) designed to execute a detection algorithm (S2) to calculate an anomaly score of the automation device (AA) based on the set of signals read in; a second processor unit (P2) designed to execute a machine localization method (S34) for locating errors in the event of an abnormality indicated by the abnormality score calculated by the first processor unit (P1), wherein the machine localization method (S34) has been trained in a training phase in order to calculate, for the calculated abnormality score, the probability of a possible cause of an error associated with the respective component (K) of the automation device based on the detected loop diagram of the automation device (AA) and to provide the probability as a result.

Description

Anomaly detection in a pneumatic system

Technical Field

The present invention relates to technical error detection and localization in a pneumatic automation device (e.g. in a production device comprising an actuator and a sensor), in particular to an error detection module, an error detection system, a method and a computer program.

Background

Basic high quality, robustness and availability requirements are placed on components or field devices in different types of automation equipment. Failure or malfunction of field devices in a process can lead to extremely high costs, especially if manufacturing breaks are caused thereby. Thus, a high degree of technical complexity is deployed in field devices in order to significantly reduce the risk of failure or in order to be able to identify and report defects independently. The functionality is repeatedly integrated in a redundant manner into the field device, permanently monitoring and verifying the measurement results internally. The need to prevent field device failures increases with the field of application (e.g., in a nuclear power plant). Thus, during the monitoring of field devices, it is necessary to ensure that the devices involved are working in an error-free manner and that any faults are detected in the most timely manner possible even before any interruption.

Such monitoring and analysis tasks use a method in the field of predictive maintenance that analyzes large amounts of sensor data from field devices. These methods are typically based on predictive maintenance algorithms. The quality of these algorithms is related to the amount of sensor data available from the system that is continuously being observed. However, these methods often fail to produce satisfactory results if only a very small amount of sensor data is available.

Furthermore, machine learning and neural network methods are known in the field of automatic decision support.

However, if an error monitoring is to be performed on a device having only a minimal sensor system (e.g. a pneumatic system having only two final position sensors), the known method cannot be used. However, there is still a need for error monitoring of these devices.

Disclosure of Invention

Starting from this, the object of the invention is to provide a method by means of which a description can be provided regarding the defects of components of an automation device, in particular of a pneumatic automation device. Thus, monitoring can be improved and the automation device as a whole can be made more reliable. At least at the component level, a description of the defect and the associated is provided.

This object is achieved by an error detection module for detecting and evaluating anomalies in an automation device, in particular a pneumatic automation device, comprising:

a read-in interface, for example a digital OPC-UA interface, for reading in digital signals from an automation device; in particular, only three digital signals may be available (the time signal of the two final position switches and the point in time of the valve switching command);

-a first processor unit designed to execute a detection algorithm to calculate an anomaly score for the automation device based on the set of read-in signals;

-a second processor unit designed to execute a machine localization method for locating errors in case the anomaly score calculated by the first processor unit indicates an anomaly, in order to calculate and provide for the anomaly score a resulting probability of a possible cause of error in relation to the respective component of the automation device. In an advantageous development, the results can even be provided in relation to sub-components of the component and thus in a more detailed manner.

The invention has the technical advantage that errors can be located directly in relation to components of the automation device and also when only a small number of sensors, in particular only two final position sensors, are installed. Thus, the error positioning is possible based on only three digital signal values (i.e., the time point of the two final position switches on the cylinder and the time point of the valve switching signal, which indicates whether the controller indicates the technical process of the valve with the command "SWITCH NOW" and thus can also be defined as a valve switching command).

To configure or train the machine positioning method, an additional (e.g. third) processor unit may be designed. The further processor unit comprises:

-a loop map read-in interface for reading in a loop map of the automation device; this is used for the purpose of training the error localization model to be generated, preferably by reading in the digitized loop map once.

In a preferred embodiment of the invention, the first processor unit (a function that can be assigned to the detection algorithm) is implemented on a device other than the second processor unit (a function with a machine localization method for locating errors with increased anomaly scores), and in particular is formed on the control unit. Thus, the system for error detection and localization can be very flexibly adapted to the corresponding hardware, and thus computationally intensive processes can be transferred to high-capacity hardware (e.g. cloud servers).

In an alternative preferred embodiment of the invention, the error detection module comprises a configuration interface as a front end for configuring and training the model. Thus, for example, a user or operator of the system can quickly and easily configure the structure of the decision tree.

In another preferred embodiment of the error detection module, it is also applicable to automation devices comprising specific architectures or typical structures. The pneumatic system comprises one to a plurality of pneumatic drives, each connected to at least one valve, wherein the plurality of valves may be arranged on one valve block and/or the plurality of valve blocks may be connected to one supply unit. Multiple actuators may also be connected to one valve at a time. The architecture is represented in an electronic circuit diagram that is read in by the system and used for computational purposes. In other exemplary embodiments, different architectures may be used as a basis. This is possible because the machine localization method takes into account the corresponding loop map and in so doing automatically recognizes the active mode and deviations from the mode and can localize possible errors by inference of the detected loop logic.

In another aspect, the invention relates to an error detection system for detecting and evaluating anomalies in an automated device, in particular a pneumatic system, comprising:

-an error detection module as described above;

gateway (to the internet, e.g. edge computer), and

-a cloud-based server connected to the error detection module through a web interface. The first processor unit and the second processor unit may be deployed (implemented and arranged) as a distributed system on different units (controllers, gateways and/or servers). It may also be formed on the same unit.

The achievement of this object has been described above in relation to a device (error detection module, system). The features, advantages, or alternative embodiments mentioned herein will also be transferred to other claimed subject matter and vice versa. In other words, the method and computer program may also be developed with features described and/or claimed in connection with a module or system. In so doing, the corresponding functional features of the method are embodied by corresponding physical modules of the system or the article of manufacture, in particular by hardware modules or microprocessor modules, and vice versa.

In a further aspect, the invention relates to a method for detecting and evaluating anomalies in an automation device, in particular a pneumatic automation device, comprising the following method steps:

-reading in (preferably two) digital signals from a (pneumatic) drive and from a digital switching command of a valve of an automation device through a read-in interface; the emitter or transmitter of the signal is also referred to as a "sensor";

-executing a detection algorithm for calculating an anomaly score for the automation device based on the set of read-in signals; in an advantageous development, the anomaly score is calculated not only for the entire automation system as a whole, but also for its individual drives in a piece-by-piece and dedicated manner. Thus, the importance can be significantly improved and provided in a more detailed manner.

-if the calculated anomaly score indicates anomaly and in particular exceeds a pre-configurable limit value: triggering a machine localization method for locating errors, wherein the machine localization method has been trained in a training phase in order to calculate the probability of possible error causes in relation to individual components or sub-components (components) of the automation device on the basis of the detected circuit diagram of the automation device for the calculated anomaly score and to provide this as a result.

In an advantageous manner, the loop map is read in from the file during the test run during the training phase in order to configure the data dependencies and dependencies. Alternatively, the loop map may also be programmed locally on the error detection module, or may be entered manually.

In an advantageous development of the invention, the machine learning method (or the second processor unit) can be designed not only to output the result with the calculated error probability for each component of the device, but also in a more detailed form, i.e. for each sub-component of the respective component. Thus, the results may be provided for the constituent parts or elements of the component in a finer granularity and in a specific manner.

In a preferred embodiment, a pattern recognition algorithm is used as the detection algorithm for calculating the anomaly score. Alternatively, the anomaly score may be calculated by accessing a memory in which the trained detection model is stored. The model may be created by an automatic classification method, in particular by a k-means algorithm. To this end, a training phase is provided in which other configurations can be created and in which the model is made to learn. The model is used to classify or distinguish between two categories, i.e., a first category having a normal reaction mode of the pneumatic system and a second category having a deviation or abnormal reaction mode. It should be noted that the detection algorithm preferably compensates for real-time signals or real-time data that occur during operation of the automation device. Thus, the detection algorithm is preferably related to the current state of the device.

In a further preferred embodiment of the invention, the signals from at least two different digital sensors and the switching signals for the valves are read in and thus represent the time points of the two final position switches on the cylinders (clamping devices) of the pneumatic system and the valve switching time points. From these three digital signals, the following four time intervals are calculated:

reaction time during cylinder extension (time interval from switching point in switching point/valve to leaving current final position);

travel time during cylinder extension (time interval from leaving one final position to reaching another final position);

reaction time during cylinder retraction (time interval from switching point in switching point/valve to leaving current final position);

travel time during cylinder retraction.

The advantage of this aspect is that based on only three digital signals (or binary signals, on/off), four specifications can be derived, which have a significant impact on error detection and optionally on error localization. Thus, error detection can also be applied to existing systems that are not yet equipped with a large number of sensors.

In an advantageous development of the invention, in addition to a minimum sensor system (with three digital signals) sufficient to execute the detection algorithm and to execute the incorrect positioning, an additional sensor can be formed in the valve, which sensor detects whether and when the valve has switched. The signal may be described as a valve switching point in time. The additional digital signal provides an additional time indication from which more detailed information can be obtained. If for example the time between "now switching valve" and "valve switched" is constant, but if also a change in reaction time is detected, the change is not due to the valve. Thus, the positioning method will indicate another possible source of error or cause.

In a further advantageous embodiment of the invention, in addition to the smallest sensor system, a pressure sensor system can also be formed on the two working connections of each valve. The pressure sensor system is implemented in, for example, applicant's motion terminal (known as VTEM) and can be used accordingly to provide more information to calculate anomaly scores and to do false positives and thereby provide more detailed positioning results. Thus, in this embodiment of the invention, the pressure signal is thus also considered as a signal for calculating the anomaly score and for error localization.

In a further advantageous embodiment of the invention, the pressure system and/or the flow sensor system can monitor a plurality of valve blocks in addition to the smallest sensor system, in order to likewise provide further information for calculating the anomaly score and for error localization and thus provide more detailed localization results.

In another preferred embodiment of the invention, after calculating the reaction time and travel time during cylinder extension and retraction, the detection algorithm performs at least one of the following processing steps:

-feature extraction; this step can reduce the amount of data. Thus, the method can be performed more quickly.

-Z-score normalization; this step is used for normalization and is related to the conversion of random variables. Thus, generalization ability and comparability can be improved. The advantage is that the physical size is scaled to a balanced standardized size;

-Principal Component Analysis (PCA); this step serves to construct and simplify a large number of data sets detected by the sensor, since a plurality of statistical variables can be approximated by a small number of most meaningful linear combinations (principal components). The calculation time can be reduced.

Classification, in particular using K-means or similar methods;

-a logic function, wherein the result of the K-means maps to a value between "0" and "1", and thus the anomaly score is normalized to a value in the interval [0, ], 1 ]; and/or

-smoothing; only at the end of the process is the corrupted sensor data smoothed. Thus, sensitivity and specificity can be adjusted.

In another preferred embodiment of the invention, the detection algorithm outputs anomaly scores in the value range [0,..1 ] and sensor correlation values as intermediate results of the method. With this intermediate result, the machine positioning method can be applied in a subsequent step.

In another preferred embodiment of the invention, the machine localization method is based on a decision tree, wherein the decision tree is calculated based on the detected loop diagram. The loop map may be read in from a file such as Eplan, fluidDraw, or from an automated ML file or a file of similar format (e.g., based on XML). Alternatively, other machine learning methods may be used. In particular, artificial neural networks for locating errors may be learned in an upstream training phase.

In a further preferred embodiment of the invention, the machine localization method extracts a data relationship between the data sets from the detected loop map and the read-in signals, wherein the data relationship is used to localize the error.

In an advantageous development of the method, the result of the machine positioning method comprises error probability values for all components of the pneumatic device, or alternatively for sub-components selected as relevant components and/or within one component. Furthermore, in other refinements, the following processing steps may be performed:

-summarizing all error probability values for all components;

-accessing a memory in which a rule system is stored for locating errors relating to individual components of the automation device.

The machine localization method comprises two phases for error localization. The first stage calculates in which part of the automation device the error is located. Thus, in the first stage, error localization is achieved at the component level. The results may be read in, for example, as follows: the "clamping device X seizes" or "the valve Y is defective". The second stage calculation error may be located at a precise location in the component. Thus, in the second stage, error localization is achieved at the sub-component level. The results may be read in, for example, as follows: "friction on cylinder", "leakage at cylinder chamber a", "hose B with leakage", "restrictor D blocked", etc.

In the machine positioning method, therefore, the probability of preferably all components (clamping device components) is first determined. In an advantageous alternative embodiment of the invention, probabilities are determined only for components determined to be relevant (e.g. in the configuration phase) in order to reduce computational resources and to be able to provide results possibly faster. It follows from this whether errors occur in the identified clamping device or whether all clamping devices of the valve are affected. If the latter is the case, a conclusion can be drawn that there is a problem on the valve by accessing the rule system. If all valves of the valve block are abnormal, the system or rule indicates that the problem is at the valve block level. Thus, by accessing the rule system, it is always possible to limit the false positioning to a specific part of the device in a finer-grained manner.

The object is also achieved by a computer program comprising a computer program code for performing all the method steps of the method described in more detail above, when the computer program is executed on a computer. In this regard, the computer program may also be stored on a computer readable medium.

The object is also achieved by a computer program product comprising a computer program code for performing all the method steps of the method described in more detail above, when the computer program is executed on a computer. A computer program product may be designed as a stored executable file, optionally comprising other components (e.g. libraries, drivers, etc.), or as an electronic unit (microprocessor, computer) comprising an installed computer program.

The terms used in this application are explained in more detail below.

The machine positioning method is a method implemented by a special computer. Machine localization methods are used to predict errors that occur in specific components of a device. To this end, a decision tree may be constructed in which a model is represented. The model may be stored in memory. The decision tree is used to operate the time to assign the object (in this case: the various components of the device, such as valves, valve blocks, compressed air sources, power sources, etc.) to the error category. In this way, probabilities can be assigned.

For example, a bayesian network or other decision logic may be applied. Basically, by observing three digital signals of the pneumatic system in relation to the operating time, the probability of component-based and sub-component-based error sources can be indicated. If a common probability distribution of a large number of variables is to be managed, resource constraints (latency, processor power, etc.) are quickly encountered for explicit representations (by indicating the probability of each state combination). For example, in the case of 20 binary variables, i.e., 20 variables each having two states, then 2 must be specified ²⁰ =1048576 individual values. By exploiting the (conditional) dependencies between the variables of the domain to be modeled, the number of values that need to be indicated can generally be reduced to a manageable size. Bayesian networks represent this approach. The bayesian random variable network consists of two parts:

1. A directed acyclic graph whose nodes correspond to random variables, and whose conditional dependencies between variables are encoded with their edges;

2. probabilities determined by the table associated with the variables.

Decision trees are built during a training or learning phase and then run through it in a top-down fashion for predictive or mislocalization purposes when in use. Neural networks or naive bayes classifiers, k-nearest neighbor methods, or support vector machines may be used as alternative techniques for the machine localization method.

The detection algorithm is a computer-implemented method for grouping or classifying data sets representing pneumatic system states (normal/abnormal) based on detected signal combinations. For this purpose, for example, a k-means algorithm can be applied. The purpose of the k-means algorithm is to divide the data set into k (in particular here 2) partitions to minimize the sum of squared deviations from the cluster centroids. In extended embodiments of the present invention, a k-median algorithm or k-means++ algorithm or similar classification algorithm may also be applied.

The read-in interface is a digital interface. For reading in digital data and in particular can operate according to the OPC unified architecture (OPC-UA) protocol. OPC-UA is an industrial machine-to-machine communication protocol that ensures interoperability. Data from a field bus, such as Profinet, can likewise be read in.

The signal is a digital signal (on/off) which can be further processed in digital form directly by the processor unit. Preferably, a digital sensor is used directly. For digital sensors, the electrical signal is directly converted digitally (a/D conversion inside the sensor). Subsequent calculations (e.g., error compensation) may be performed in the microprocessor. Alternatively, an analog sensor may also be provided, the analog signal of which is converted into a digital signal in an external or separate a/D converter. The digital signal may then be used as a digital value and may be output via any digital protocol (e.g., USB, CANopen, or Profibus). During further transmission, the digital pressure signal is not affected by disturbances, which may lead to a reduced accuracy.

The error detection module is an electronic module which can be distributed over several components and is designed with the function of avoiding errors and of positioning the components of the pneumatic device in error. In particular, an error detection module, which may be implemented locally on a device of an automation apparatus, may be executed centrally with access, and in particular cloud-based computing. The error detection module is arranged to implement control measures and/or diagnostic measures if the maintenance software identifies a possible failure of a component of the automation device in time. Defective components that may quickly lead to equipment downtime can thus be identified independently of general maintenance time and replaced before damage actually occurs. This allows to achieve cost savings in terms of daily maintenance or time-dependent preventive maintenance, since work can only be done when actually needed. Within the scope of the invention, the analysis is preferably performed in parallel with the operation of the device, to avoid downtime.

A gateway (node) is a computer-based unit that can be designed as an edge computer close to the site and has a cloud-based interface (web interface) to the server. The gateway calculates the anomaly score and processes it further as part of the error localization. The results may be communicated to a server and/or at a field level (e.g., programmable logic controller).

The component is a field device and therefore a technical device which is directly related to the production process in the field of automation technology. In automation technology, the term "field" refers to an area outside a control cabinet or control room. Thus, field devices can be actuators (control elements, valves, etc.) and sensors (measuring transducers) in plant and process automation. These components are connected to the control and management system mainly by means of a field bus. The component can be designed with sensors in order to detect, generate or aggregate sensor data, so that the data can be used in an evaluated manner for regulation, control and further processing. These components are part of an automated apparatus that may include devices (e.g., an industrial robot).

The control device is an electronic module for controlling (open loop control) and/or regulating (closed loop control) a machine or automation device having a set of field devices and is programmed digitally. In particular, they may be programmable logic controllers. In the simplest case, the control device has an input, an output, an operating system (firmware) and an interface through which the user program can be loaded. The user program determines how to switch the output based on the input. The operating system ensures that the current state of the sender is always available to the user program. Based on this information, the user program can switch the output to cause the machine or automation device to operate in a desired manner. The control device is connected to the automation device and to its field devices via sensors and actuators.

In the following detailed description of the drawings, discussion of the drawings will be understood as non-limiting exemplary embodiments and their features and other advantages.

Drawings

In the following detailed description of the drawings, discussion will be understood as non-limiting exemplary embodiments and their features and other advantages by means of the accompanying drawings. In the drawings:

FIG. 1 shows an overview of an error detection system of the present invention including an error detection module;

FIG. 2 shows an embodiment of an error detection module, which is an alternative to that illustrated in FIG. 1;

FIG. 3 illustrates another schematic diagram of an error detection module including a cloud-based server and other components;

FIG. 4 shows an alternative schematic diagram design of an error detection module;

FIG. 5 shows a flow chart of method steps of an error detection method according to a preferred embodiment of the invention, and

FIG. 6 shows a schematic diagram of an error detection system including other components in accordance with a preferred embodiment of the present invention.

Detailed Description

The present invention is technically used for monitoring pneumatic systems as an example of an automation system or device comprising various field devices (also referred to as components in the following) which are controlled by a control device, for example a programmable logic controller. In particular, errors should be identified in time, and preferably at some point in time before the individual components fail or cause errors in the device. For this purpose, an error detection module will be used, which will be explained in more detail below with respect to fig. 1.

The invention has the advantage that early error detection is possible for complex, multi-part, preferably pneumatic, automated devices, although only a small amount of measurement data is available and can be operated similarly with a minimum of sensor systems. In particular, although only two digital sensors and one switching command are used, in particular one sensor for detecting the time points of the two final position sensors on one cylinder and for detecting the valve switching time point, it is possible to provide a result of erroneous positioning. This has the advantage that abnormality detection is also possible in devices in which only the actuator is equipped with a sensor system (e.g. a final position sensor). The method presented here is based on a model in which at least these signals are taken into account. Optionally, other signals are also contemplated, such as pressure signals and/or flow signals or other signals of sensors inside the valve, which are detected in the pressure source and/or the valve. By means of the detection algorithm, deviations or changes from the correct or typical reaction behavior of the pneumatic device can now be detected automatically and in real time, for example, the time between "valve switching" and "leaving final position 1" and the travel time (final position 1 to final position 2). Furthermore, in principle the time between the physical switching of the sending of the control command to the valve is measured and known. In an advantageous development, an additional valve internal sensor can be provided, which detects whether the valve has been switched. The same applies to the return movement of the valve. The measured variable and the pattern derived from the measured variable are obtained during "good" operation (i.e., during error-free operation). The error image shows a characteristic pattern for anomaly detection and error localization according to the present invention. Furthermore, the circuit diagram of the pneumatic system can be used in a digital pneumatic circuit diagram, which can be read in, for example, from a Fluid Draw or Eplan or an automated ML file, and used to construct the decision logic. If a deviation from a good pattern is detected by means of a detection algorithm, a false localization can be provided in a second step by applying a machine localization method. Logic circuits including implemented decision logic may be used for this purpose, for example, using decision trees or bayesian networks or other machine learning methods.

The background of the proposed solution in this case is: the time behaviour of a tensioning or clamping system (e.g. car, body) consisting of valve, hose system and clamping device changes with increasing wear. Test arrangements are created to identify whether and how the manipulations performed on the pneumatic system affect the temporal behavior. Pneumatic systems have been varied and manipulated with a targeted view. This includes friction and leakage at the clamp and valve, as well as variations in lever arm length, hose length between the valve and clamp, and supply pressure. The closing time and delay time have been recorded when the cylinder was opened and closed. As a result of tests carried out by the applicant, it can be said that the friction, leakage and variations in the supply pressure of the clamping device affect the delay and closing times which can be derived from the final position switch signal. The result of the test arrangement influences the configuration of the error localization model, wherein in a first stage errors are localized with respect to the individual parts of the device and in a second stage errors are localized with respect to the individual sub-parts of the parts. It is possible to clearly identify which type of malfunction is present. Thus, errors and in particular positioning errors can be contained based on the (three) digital signals.

Fig. 1 schematically shows an error detection module FM. It comprises on one side of the automation device AA (preferably pneumatic) components, for example valve blocks or valve disks, wherein the valve blocks in turn comprise valves with clamping means/cylinder units and/or also pneumatic actuators (e.g. pneumatic drives etc.) and sensors and pressure sources. Furthermore, a controller is provided, which may be designed as a programmable logic controller, which may also be referred to as a PLC. The component is designed with a sensor for detecting a digital signal or a switching command to the valve. The first part comprises at least one sensor unit S1 for detecting three digital signals, the second part comprises in turn a sensor unit S2 for detecting at least three signals, etc.

As shown in fig. 1, the further sensor S3 may also send a signal (e.g. a pressure signal) to the programmable logic controller. The controller receives digital signals via the read-in interface I1 and is furthermore designed with a first processor unit P1 for executing a detection algorithm on the basis of the detected or read-in signals. The detection algorithm is used to calculate an anomaly score for the automation device AA based on the detected or read-in signal set. The calculated anomaly score may be communicated to the IoT gateway GW via a data interface (e.g., OPC-UA). The calculated anomaly score and/or the detected signal is transmitted via the second interface I2 to the second processor unit P2, which second processor unit P2 may be designed to perform a machine localization method S34 (described in more detail below with reference to fig. 5) for locating errors if the anomaly score calculated using the first processor unit indicates an anomaly, in order to calculate and provide for the anomaly score a result probability of a possible cause of an error in relation to the respective component of the automation device AA.

In the example shown in fig. 1, the first processor unit P1 is implemented on a different device than the second processor unit P2. The first processor unit P1 may be formed on the control unit and the second processor unit P2 may be formed, for example, on a gateway node GW (or gateway for short). To perform the machine positioning method, the second processor unit P2 accesses a memory MEM in which the trained model is stored. The second processor unit P2 receives the circuit diagram of the pneumatic device AA via the circuit diagram reading interface I3. The circuit diagram is provided in digital form and contains information about the structure of the device AA and about the function (in particular the switching point in time of the valve, etc.).

In the exemplary embodiment shown in fig. 1, a separate gateway GW is provided, which gateway GW serves as an intermediary between the device AA with components and the programmable logic controller on the one hand and as a server SV on the other hand. The gateway GW may be implemented, for example, in a superior management system of the device AA and/or may be assigned to the device AA (e.g., in the same security domain as the device). The third processor unit P3 may be formed on the server SV in order to be able to perform a machine positioning method on a cloud-based server, for example.

As schematically shown in fig. 1, basically, the first processor unit P1 may send the locally calculated anomaly score similarly as an intermediate result to the second processor unit P2 (solid arrow). Alternatively or additionally, the detected signal may also be transmitted to the second processor unit P2. This may be accomplished directly from the sensor and/or from the component (both shown in phantom in fig. 1) and/or from the controller.

Fig. 2 shows an alternative embodiment, wherein the gateway GW comprises a second processor unit P2 and a first processor unit P1. The component sends its three digital signals to the controller, which then sends the signals to the second processor unit P2 via a network connection (second interface I2). Alternatively, the component may send the locally detected signal directly to the second processor unit P2 (without bypassing by the controller). It is even possible that the sensor itself may be designed with an additional network interface in order to transmit data.

Fig. 3 shows an exemplary embodiment using a cloud-based server SV. The sensor data is then detected on the components of the pneumatic device AA. The first processor unit P1 can now be formed either locally in the controller or on one of the IoT gateway nodes GW assigned to the device and which can be designed as an edge computer. The gateway GW exchanges data with the server SV via an internet protocol based data connection (e.g., https, etc.), on which a second processor unit P2 designed to perform a machine positioning method is formed. The learned model may be stored in a memory MEM of the server SV. Thus, higher computing resources (and memory resources) of the server may be used to locate errors and computation results.

As the above examples anticipate, the functionality of the error detection module FM may also be implemented in a distributed manner for two aspects:

1. detection algorithm S2

2. Machine positioning method S34.

In other words, the first processor unit P1 and the second processor unit P2 may be implemented on different computer-based entities. Additional processor units may also be designed for configuring a model or training a positioning method based on training data. The training data may comprise patterns of signal combinations in good conditions (error free operation of the device).

As shown in fig. 4, the detection algorithm S2 is preferably executed as locally as possible in the controller, preferably in the vicinity of the generated data, and the machine positioning method S34 may be executed on an entity providing sufficient resources, preferably on the server SV. Only one client for model checking of the machine positioning method S34 may then be installed on the gateway GW and thus a computationally intensive process may be performed on the server SV and the results output only to the configurable entity, in particular to the gateway GW and optionally to components of the device AA and/or to the controller. The output may be realized through an output interface AS.

Fig. 5 shows a flow chart of an error detection method. After the start, a digital signal is read in step S1. In step S2, a detection algorithm is performed on or with the read-in signal. It calculates the anomaly score and the sensor correlation value as intermediate results. Thus, the intermediate result indicates whether there is an abnormality in the apparatus AA. Depending on the result, the method branches to different calculation cases, as shown in fig. 5. If no anomalies are present, the device appears to operate "as before" -i.e., in an error-free manner. The method may end or restart through EXIT. Otherwise (when an anomaly or deviation is detected), a machine localization method that has been trained in the training phase is performed in step 34 in order to calculate the probability of a possible error cause for the respectively calculated anomaly score based on a digitally or manually detected loop diagram of the automation device AA. The machine positioning method may include two stages. In a first step S3, the localization of errors (e.g. errors in the clamping device X or the valve Y) is calculated at the component level, and in a second step S4, the localization of errors is calculated at the sub-component level. In a second step S4, the analysis may locate a location in the component where an error is identified as defective. The machine localization method may be implemented as an algorithm whose execution takes into account the information of the detected loop map (design, architecture and structure of the loop and the switching point in time). As mentioned above, the functions of the algorithm may also be implemented on other devices or servers SV.

Fig. 6 is a further constructional architecture diagram of an error detection system comprising a first processor unit P1, which first processor unit P1 is in this case implemented on a processor, and comprising a second processor unit P2 implemented on a server SV exchanging data with a gateway GW over a data connection. Furthermore, a configuration interface Config-UI may be provided, through which the machine positioning method, in particular the algorithms S3, S4, may be configured. The configuration interface Contis-UI is preferably cloud-based or may be provided locally as a computer program. The configuration interface Config-UI may comprise user interface elements, such as a dashboard. In this case, it is also possible to install a version of a learned model (e.g., a constructed decision tree) with a training master as an application for configuring the learning stage of the model or generating the decision tree, and a score master as an application for calculating anomaly scores according to other options. A set of applications for error detection and localization may be installed on a server SV, such as an industrial PC. In particular, a runtime environment (e.g., java runtime environment) of the trained model is implemented that is synchronized with the configuration interface Config-UI and interacts with the gateway GW, preferably through https/REST-Uload requests. The read-in signals are then sent to the server through the gateway GW for error detection and localization purposes.

In one exemplary embodiment, another processor unit, defined as third processor unit P3 in fig. 6, may be provided and used to generate a model for machine positioning method S34. The user has the option to adjust the settings through the configuration interface Config-UI. The functions for generating the model may also be implemented on the server SV.

In this exemplary embodiment, the IoT gateway node GW may be designed with a client for the machine positioning method. The client/gateway may be located in the field near the device. The gateway GW may have a browsing function that may be used to browse page by page and check anomaly scores transmitted by the controller. Further, the gateway node GW may have a proxy for the algorithm provided thereon, which proxy may operate in the cloud (e.g. on the server SV); and may have agents for automation suites with other applications and programs as PC applications. The automation suite functions the same as the cloud. Further, the gateway GW may have a circular buffer for intermediately storing data formed thereon; and a condensed version of the trained model (for performing machine location methods) for persistence, configuration, license management, and other functions in conjunction with the machine location methods. Fundamentally, depending on the configuration, the gateway GW may have other programs installed thereon, which may also run in the background and may provide specific services, among other things. The user interaction preferably takes place only indirectly, for example via signals, pipes and especially (network) sockets.

In one test, 6 pneumatic clamps were operated continuously for a long period of time until wear occurred, with an extended run time compared to normal operation, or a reduced cycle time compared to normal operation. The signs of wear were seen in all the clamping device data 2 weeks before failure. According to the invention, faults and induced error conditions can be detected by machine localization methods or trained models, and automatic process monitoring can be performed.

Finally, it is noted that the description and exemplary embodiments of the present invention should be construed in a radical sense as being non-limiting with respect to the particular physical embodiment of the invention. All the features explained and illustrated in connection with the various embodiments of the invention can be provided in different combinations in the subject matter according to the invention in order to achieve their advantageous effects at the same time.

The scope of the invention is defined by the appended claims and is not limited by the features explained in the description or shown in the drawings.

It is particularly obvious to a person skilled in the art that the invention can be used not only for pneumatic devices, but also for other hydraulic devices or other fluid technology systems or electric spindles. Furthermore, the components of the error detection module may be distributed across multiple physical products.

Claims

1. An error detection module (FM) for detecting and evaluating anomalies in an automation device (AA), comprising:

-a read-in interface (I1) for reading in digital signals from the automation device (AA), wherein the digital signals comprise at least two digital signals from at least two different digital sensors and a digital signal for a switching command of a valve;

-a first processor unit (P1) designed to execute a detection algorithm (S2) to calculate an anomaly score of the automation device (AA) based on the digital signal read in from the automation device;

-a second processor unit (P2) designed to execute a machine localization method (S34) for locating errors in case the anomaly score calculated with the first processor unit (P1) indicates anomalies, wherein the machine localization method (S34) has been trained in a training phase in order to calculate the probability of a possible cause of an error in relation to the individual components of the automation device based on the detected loop diagram of the automation device (AA) for the calculated anomaly score and to provide this probability as a result.

2. The error detection module (FM) according to claim 1, wherein the first processor unit (P1) is implemented on a different device than the second processor unit (P2).

3. The error detection module (FM) according to claim 1, wherein the first processor unit (P1) is implemented on a control unit.

4. An error detection module (FM) according to any of claims 1-3, wherein the second processor unit (P2) or another processor unit (P3) designed to generate an error localization model comprises a loop-map reading-in interface (I3) for digitally reading in a loop-map of the automation device (AA).

5. The error detection module (FM) according to claim 4, comprising a configuration interface (Config-UI) as a front end for configuring and/or training the error localization model.

6. An error detection module (FM) according to any of claims 1-3, wherein the automation device (AA) comprises a pneumatic system with a pneumatic drive, wherein a plurality of drives and/or actuators are connected to valves and a plurality of valves are arranged on a valve block, and a plurality of valve blocks are connected to a supply unit.

7. The error detection module (FM) of claim 1, wherein the automation device is a pneumatic automation device.

8. The error detection module (FM) of claim 7, wherein the digital signal is representative of a point in time of two final position switches on a cylinder of the pneumatic automation device and a valve switching point in time.

9. The error detection module (FM) according to claim 8, wherein four time intervals are calculated from the at least two digital signals and the digital signal for the switching command of the valve:

-a reaction time during extension of the cylinder;

-a travel time during extension of the cylinder;

-a reaction time during retraction of the cylinder;

-a travel time during retraction of the cylinder.

10. A method for detecting and evaluating anomalies in an automation device (AA), comprising the following method steps:

-reading (S1) digital signals of the automation device (AA) via a reading-in interface (I1), wherein the digital signals comprise at least two digital signals from at least two different digital sensors and a digital signal for a switching command of a valve;

-executing a detection algorithm (S2) to calculate an anomaly score for the automation device based on the read-in digital signal of the automation device;

-if the calculated anomaly score indicates anomaly, then: triggering a machine localization method (S34) to localize errors, wherein the machine localization method (S34) has been trained during a training phase in order to calculate, for the calculated anomaly score, a probability of a possible cause of error associated with the individual components of the automation device based on the detected loop diagram of the automation device (AA) and to provide the probability as a result.

11. The method according to claim 10, wherein the detection algorithm (S2) for calculating the anomaly score is a pattern recognition algorithm or is implemented by accessing a memory storing a trained detection model.

12. The method according to any of claims 10-11, wherein the machine localization method (S34) calculates (S4) a probability of a possible error cause associated with each sub-component of a component.

13. The method of claim 10, wherein the automated device is a pneumatic automated device.

14. The method of claim 13, wherein the digital signal represents a point in time of two final position switches on a cylinder of the pneumatic automation device and a valve switching point in time.

15. The method of claim 14, wherein four time intervals are calculated from the at least two digital signals and the digital signal for the switching command of the valve:

-a reaction time during extension of the cylinder;

-a travel time during extension of the cylinder;

-a reaction time during retraction of the cylinder;

-a travel time during retraction of the cylinder.

16. Method according to claim 14, wherein signals of two final position switches are read in, said signals comprising a valve switching time point signal and/or a pressure signal and/or a flow signal.

17. Method according to claim 15, wherein after calculating the reaction time and travel time during extension and retraction of the cylinder, the detection algorithm (S2) performs the following processing steps:

-feature extraction;

-Z-score normalization;

-principal component analysis;

-classification using K-means;

-a logic function; and/or

-smoothing.

18. The method according to any one of claims 10, 11 and 13-17, wherein the detection algorithm (S2) comprises as a result an anomaly score and a sensor correlation value.

19. The method according to any of claims 10, 11 and 13-17, wherein the machine localization method (S34) comprises a decision tree method, wherein the decision tree is calculated based on a detected loop diagram or comprises a bayesian network method.

20. The method according to any of claims 10, 11 and 13-17, wherein the machine localization method (S34) extracts data relationships between data sets based on read-in signals from the detected loop map.

21. The method according to any of claims 10, 11 and 13-17, wherein the result of the machine positioning method (S34) comprises error probability values for all components and/or all sub-components of a component, and wherein the method further performs the following processing steps:

-summarizing all error probability values;

-accessing a memory in which a rule system is stored for locating errors relating to individual components and/or sub-components of the automation device (AA).

22. An error detection system for detecting and evaluating anomalies in an Automated Apparatus (AA), comprising:

-an error detection module (FM) according to any of claims 1-9 related to an error detection module;

-Gateway (GW)

-a cloud-based Server (SV) connected to said error detection module (FM) through a web interface.

23. The error detection system of claim 22, wherein the automated device is a pneumatic automated device.

24. A computer program product comprising a computer program, wherein the computer program performs all the method steps of the method according to any of the preceding method claims 10-21 when said computer program is executed on a computer.